Prediction of opioid use disorder among patients with chronic non-cancer pain receiving long-term opioid therapy.

OBJECTIVE

Assessing opioid use disorder risk in patients prescribed long-term opioid therapy for management of chronic non-cancer pain is critical for prevention and early intervention.

DESIGN

Case-control study.

SETTING

Pain management and primary care clinics, and substance use treatment facilities.

SUBJECTS: Participants are 1300 patients with chronic non-cancer pain (59.68% women; mean age = 49.03 years), 409 of whom developed opioid use disorder.

METHODS

We compared the performance of 3 machine learning models that used the Opioid Risk Tool for Opioid Use Disorder alone with those that incorporated an expanded set of clinical predictors.

RESULTS

The Opioid Risk Tool for Opioid Use Disorder showed strong performance (precision = 0.91; specificity = 0.96). Models that incorporated additional predictors showed improved performance on precision-recall area under the curve and F1 scores, particularly the random forest and eXtreme Gradient Boosting models. Aside from the Opioid Risk Tool for Opioid Use Disorder, the most important features in the expanded models were nicotine dependence, marital status, opioid misuse behaviors, and pain interference and catastrophizing.

CONCLUSIONS

A stepwise approach that employs the Opioid Risk Tool for Opioid Use Disorder as a preliminary screener followed by a more in-depth assessment of clinical predictors among high-risk individuals may offer a feasible strategy to optimize efficiency and precision in risk stratification. Future work should refine and validate this framework in diverse population and care settings, as well as examine its integration into clinical workflow to enhance the identification of chronic non-cancer pain patients at risk for opioid use disorder.